ansible-test - Improve container probe error handling (#84109)

3 files changed, 28 insertions, 9 deletions

diff --git a/changelogs/fragments/ansible-test-probe-error-handling.yml b/changelogs/fragments/ansible-test-probe-error-handling.yml
new file mode 100644
index 0000000000..bf4301cc48
--- /dev/null
+++ b/changelogs/fragments/ansible-test-probe-error-handling.yml
@@ -0,0 +1,3 @@
+minor_changes:
+  - ansible-test - Improve container runtime probe error handling.
+    When unexpected probe output is encountered, an error with more useful debugging information is provided.
diff --git a/test/lib/ansible_test/_internal/docker_util.py b/test/lib/ansible_test/_internal/docker_util.py
index 6bdcf927b7..0fd22fbcec 100644
--- a/test/lib/ansible_test/_internal/docker_util.py
+++ b/test/lib/ansible_test/_internal/docker_util.py
@@ -20,6 +20,8 @@ from .util import (
     SubprocessError,
     cache,
     OutputStream,
+    InternalError,
+    format_command_output,
 )
 
 from .util_common import (
@@ -300,7 +302,7 @@ def detect_host_properties(args: CommonConfig) -> ContainerHostProperties:
     options = ['--volume', '/sys/fs/cgroup:/probe:ro']
     cmd = ['sh', '-c', ' && echo "-" && '.join(multi_line_commands)]
 
-    stdout = run_utility_container(args, 'ansible-test-probe', cmd, options)[0]
+    stdout, stderr = run_utility_container(args, 'ansible-test-probe', cmd, options)
 
     if args.explain:
         return ContainerHostProperties(
@@ -313,6 +315,12 @@ def detect_host_properties(args: CommonConfig) -> ContainerHostProperties:
 
     blocks = stdout.split('\n-\n')
 
+    if len(blocks) != len(multi_line_commands):
+        message = f'Unexpected probe output. Expected {len(multi_line_commands)} blocks but found {len(blocks)}.\n'
+        message += format_command_output(stdout, stderr)
+
+        raise InternalError(message.strip())
+
     values = blocks[0].split('\n')
 
     audit_parts = values[0].split(' ', 1)
diff --git a/test/lib/ansible_test/_internal/util.py b/test/lib/ansible_test/_internal/util.py
index 23d6a81209..b83f3d5db7 100644
--- a/test/lib/ansible_test/_internal/util.py
+++ b/test/lib/ansible_test/_internal/util.py
@@ -970,14 +970,7 @@ class SubprocessError(ApplicationError):
         error_callback: t.Optional[c.Callable[[SubprocessError], None]] = None,
     ) -> None:
         message = 'Command "%s" returned exit status %s.\n' % (shlex.join(cmd), status)
-
-        if stderr:
-            message += '>>> Standard Error\n'
-            message += '%s%s\n' % (stderr.strip(), Display.clear)
-
-        if stdout:
-            message += '>>> Standard Output\n'
-            message += '%s%s\n' % (stdout.strip(), Display.clear)
+        message += format_command_output(stdout, stderr)
 
         self.cmd = cmd
         self.message = message
@@ -1021,6 +1014,21 @@ class HostConnectionError(ApplicationError):
             self._callback()
 
 
+def format_command_output(stdout: str, stderr: str) -> str:
+    """Return a formatted string containing the given stdout and stderr (if any)."""
+    message = ''
+
+    if stderr := stderr.strip():
+        message += '>>> Standard Error\n'
+        message += f'{stderr}{Display.clear}\n'
+
+    if stdout := stdout.strip():
+        message += '>>> Standard Output\n'
+        message += f'{stdout}{Display.clear}\n'
+
+    return message
+
+
 def retry(func: t.Callable[..., TValue], ex_type: t.Type[BaseException] = SubprocessError, sleep: int = 10, attempts: int = 10, warn: bool = True) -> TValue:
     """Retry the specified function on failure."""
     for dummy in range(1, attempts):